Aug 17, 2026 Technology Pulse

📰 I stopped my Claude Code subagents from running on Fable instead of Sonnet

Thomas Witt described how he stopped Claude Code subagents from running on Fable instead of Sonnet. The post is on Hacker News with two points and no comments. Read full analysis →

🧠 Cutting RAG inference costs 6x starts with deciding what never reaches the LLM

Most teams building retrieval-augmented generation systems for high-stakes classification route every ambiguous case straight to the language model and trust the retrieved context to sort it out. VentureBeat reports that this works in a demo but falls apart when the system has to survive an audit, a regulator, or compliance. Cutting RAG inference costs 6x, the piece argues, starts with deciding what never reaches the LLM. Read full analysis →

🛡️ An open agent-security benchmark, including the attacks we fail to catch

An open agent-security benchmark covering attacks current defenses fail to catch has been posted on GitHub at github.com/AndrewSispoidis/contemporary-agent-attacks. The listing appeared on Hacker News under HN AI Agents with one point and no comments. The discussion thread is at news.ycombinator.com/item?id=49320527. Read full analysis →

📰 Show HN: Graft – Claude Code hooks that cut grep tokens by 42%

Graft is a Show HN project of Claude Code hooks that cut grep tokens by 42 percent. The code is on GitHub under NanoNets/Graft, and the Hacker News thread has 34 points and 32 comments. Read full analysis →

📰 The Agent Stack: From Keyboard to Model

A Hacker News AI Agents thread linked to The Agent Stack: From Keyboard to Model at the-agent-stack.com. The submission had 4 points and 0 comments. Read full analysis →

🧠 Agent-kit – a size-based mandatory review chain for Claude Code

Agent-kit is a size-based mandatory review chain for Claude Code, published on GitHub at github.com/horilla98/agent-kit. The project was shared on Hacker News in the Claude, Codex, and Fable discussion, where it had one point and no comments. Read full analysis →

🛡️ Ironic-Python-Agent: Container HardwareManager Security Model Misimplemented

Ironic-Python-Agent has a misimplemented Container HardwareManager security model, according to a report shared via HN AI Agents. The write-up is posted on oss-sec, and the Hacker News thread has two points and no comments. Read full analysis →

📰 Issues Reaching Status.claude.com

Users are reporting issues reaching Status.claude.com. Anthropic has an open incident for the problem on its status page. A Hacker News thread on the report has 3 points and no comments. Read full analysis →

📰 Why Does Claude.md Keep Growing? Catastrophic Remembering in Agentic Coding

A Hacker News item titled Why Does Claude.md Keep Growing? Catastrophic Remembering in Agentic Coding links to an arXiv paper at https://arxiv.org/abs/2608.11095. The thread, listed under HN Claude/Codex/Fable, currently has 1 point and 1 comment. The paper addresses catastrophic remembering in agentic coding. Read full analysis →

📰 Is HTTPS://status.claude.com loading for you all?

Hacker News users are asking whether https://status.claude.com is loading. The thread has 7 points and 7 comments at https://news.ycombinator.com/item?id=49295148. Read full analysis →

🔎 More Interesting News

• 9to5Mac Daily: August 13, 2026 – MacBook Air shortages, iPhone Ultra launch rumors — 9to5Mac
• GLM 5.2 free for eve agents through August 27 via Blackbox on AI Gateway — Vercel Blog
• Databricks wanted to raise $1B, investors wanted $15B. It settled on $5B at a $190B valuation. — TechCrunch
• New HomePod mini 2: Release date, features, price, and rumors — 9to5Mac
• Hackers breach govt webmail while running parallel crypto fraud — BleepingComputer
• AI 'watermark removers' flood the web. Almost none can prove they work. — BleepingComputer
• Vercel Launches v0 API for Headless App Building — InfoQ